TL;DR I will post and track my progress in developing a new L/S equity strategy in this thread.
In the excellent tearsheet review webinar last week, I heard Jess' message on the pitfalls of overfitting loud and clear, and similar to others (@Olive Coyote, @Leo M., etc) I've decided to take this fully onboard.
Therefore, rather than trying to reduce the likelihood of overfitting in my original PARTY algo (which I may come back to at a later date), I've decided to start from scratch on a new strategy with a different (albeit with some similarities) economic hypothesis. I plan to track progress of my development in this thread. Feel free to post any constructive critical feedback, words of encouragement, or just stop by and say 'Hi.' :o)
Primary Objective: To minimize the likelihood of overfitting in a from-the-ground-up developed strategy that meets all of Q's contest criteria.
- Scalable strategy in terms of position concentration: >400 positions
each side (minimum)
- Scalable strategy in terms of capital deployed: >10MM (minimum)
- Daily Turnover: 5% < Average & Max > 20% (the closer to 5% the
- Minimum Volatility: As close to 2% (annualized) as possible, but not
- Beta to SPY: As close to 0 as possible
- Beta to Peers: As close to 0 as possible (currently impossible to
- Sector Exposure: Minimal individual sector exposure before
- Style Exposure: Minimal individual style exposure before constraining (aspirational)
Requirements to achieve Objectives:
Only In-sample data will be used in researching and discovering alpha factors (data holdback).
No back-testing will be done until alpha factors have been completely researched and understood.
In-sample timeseries selected: 06-15-2015 to 07-15-2017 (09-15-2017 including 2 months of future returns)
1st OOS AL test period: 06-15-2007 to 07-15-2009 (09-15-2009 including 2 months of future returns)
2nd OOS AL test period: TBD
Final OOS time-series selected (for final backtest): 07-16-2017 to Present
Trading costs: Q's default commission and slippage
Strategy Development Life Cycle:
- Economic Hypothesis - Completed (GARP)
- Data Selection - Completed (Morningstar Fundamentals & OHLCV)
- Universe Definition - Completed (Base Universe = QTradableStocksUS)
- Alpha Factor Research & Discovery - In Progress
- Alpha Factor Combination
- Risk Constraints Selection
- Strategy Backtesting
- Backtest Analysis
A few notes on the above:
- #4 will be done exclusively in the Research environment.
- I anticipate spending most of my time on #4, using mostly Alphalens. This may take weeks, and completion of the strategy may take months.
- The selected In-sample time-series will be used to train my initial models in #4.
- The selected OOS time-series, for testing my trained model in AL, will become part of the In-Sample training period after it's been used to test a model OOS.
- The final OOS time-series (for backtesting) will be used first in #7, which will include both the selected In-sample and OOS time-series.
- The final OOS time-series selected (for backtesting) will also be used in #8 with Pyfolio bt option set as (live_start_date = '07-16-2017', hide_positions = True, round_trips = True).
- I realize the workflow I've chosen is simpler, less 'Agile,' and more linear than 'A Professional Quant Equity Workflow' and 'Idea to Algorithm' but I've tried to base it on both. Hopefully it's still not too prone to 'overfitting,' which is what I'm primarily trying to avoid.
- I don't anticipate attaching full Alphalens notebooks, so not to disclose any discovered alpha, but I plan to attach excerpts and my analysis and decisions based on the output.
- No conscious efforts will be done to 'maximize' Returns or rolling 6mo sharpe ratio, other than researching, understanding, and selecting alpha factors based on analysis done in the Research environment.
- The 'Secondary Objectives' are just that - secondary. They are more 'nice-to-haves' and are fine as long as they don't interfere with the Primary Objective.
- This post may be edited to reflect any feedback from the community, and what I may learn during the strategy development process as I go along.
- A disclaimer: If you ask questions to me in this thread, I may choose to not answer them; nothing personal if I don't. :)
A brief description of my (1.) Economic Hypothesis (GARP) to follow...
EDITS: 21-08-2018, I've made a few updates, mostly regarding the OOS period selected for AL research vs OOS for the final backtest.